StreamKL is the first fused GPU primitive for attention KL divergence that reduces memory from O(N_Q N_K) to O(1) via an online one-pass formulation and tile-wise recomputation.
super hub Mixed citations
DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models
Mixed citation behavior. Most common role is background (55%).
344
Pith papers citing it
279
external citations · Pith
Background
55% of classified citations
abstract
We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. The key technical breakthroughs of DeepSeek-V3.2 are as follows: (1) DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 and exhibits reasoning proficiency on par with Gemini-3.0-Pro, achieving gold-medal performance in both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI). (3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.
hub tools
citation-role summary
background 31
baseline 17
method 5
dataset 2
other 1
citation-polarity summary
claims ledger
- abstract We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. The key technical breakthroughs of DeepSeek-V3.2 are as follows: (1) DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-com
authors
co-cited works
representative citing papers
LoHoSearch is a new benchmark of 544 KG-constructed questions across 11 domains where the strongest search agent scores 34.74% and context strategies add at most 6.8%.
LITMUS is the first benchmark using semantic-physical dual verification and OS state rollback to measure behavioral jailbreaks in LLM agents, revealing that even strong models execute 40%+ of high-risk operations and exhibit execution hallucination.
Agent-ValueBench is the first dedicated benchmark for agent values, showing they diverge from LLM values, form a homogeneous 'Value Tide' across models, and bend under harnesses and skill steering.
ReLibra uses pre-known token-to-expert routing from RL rollouts to perform inter-batch expert reordering and intra-batch replication, delivering up to 1.6x higher throughput than Megatron-LM and 1.2x over oracle-equipped EPLB while staying within 6-10% of an ideal balanced baseline.
Harmful skills in open agent ecosystems raise average harm scores from 0.27 to 0.76 across six LLMs by lowering refusal rates when tasks are presented via pre-installed skills.
HWE-Bench is the first repository-level benchmark for LLM agents on real hardware bug repair, where the best agent fixes 70.7% of 417 tasks but drops below 65% on complex SoC projects.
OccuBench is a new benchmark for AI agents on real-world occupational tasks via LLM-driven simulators, showing no model dominates all industries, implicit faults are hardest, and larger models with more reasoning perform better.
AlpsBench supplies 2500 real-dialogue sequences with verified memories to benchmark LLM extraction, updating, retrieval, and utilization of personalized information.
EHR-derived standardized patients and dual-track evaluation reveal LLMs trail clinicians by 37.28 points on full psychiatric encounters, with mental-status assessment the main bottleneck.
Hallucination Self-Play co-evolves a generator and detector from one base LLM via RLAIF and RLVR, lifting a 7B model to match advanced LLMs on RAGTruth faithfulness detection.
Lynx partitions KV cache bits into anchor and residual streams for progressive transfer, enabling speculative decoding on partial data followed by verification to match BF16 accuracy at 4-bit-like TTFT.
Static SFT and RL training for tool-use agents leads to performance drops under open-world distributional shifts across perception, interaction, reasoning and internalization; perturbation-augmented fine-tuning is proposed as mitigation.
Self-GC governs agent context as indexed objects with planner-proposed actions, achieving 84.85% no-impact on future continuations on a hard set versus 54-70% for baselines.
SmoothAgent introduces lookahead context engineering to eliminate transformation overhead in LLM agents, reducing TTFT by up to 11.9x through proactive KV cache preparation.
Hiding explicit demographic labels raises harmful LLM decisions by +4.4 pp even when models correctly recover the identity, exposing performative rather than genuine moral safety.
OmniCoT is a new panoramic reasoning benchmark with 6.7K eval, 1K real, and 14.3K training examples plus a two-stage SFT+GRPO training method to enforce global 360-degree consistency.
Cortex uses an Ontological Corpus Graph to structure web-scale corpora, creating a refined 24.14B-token corpus and a new benchmark validated on eight LLMs.
SpreadsheetBench 2 provides 321 expert-validated tasks from authentic business data showing frontier LLMs reach only 34.89% overall accuracy on end-to-end spreadsheet workflows.
Dockerless uses agentic repository exploration to verify patches without execution, enabling SFT and RL training of coding agents that reach 62.0/50.0/35.2% resolve rates on SWE-bench Verified/Multilingual/Pro while matching environment-based results.
BigBag generates reusable AST transformation rules via LLMs that achieve 78.6% fix rate on 157 breaking dependency updates and 33.3% cross-project transfer overall.
BehaviorBench is a benchmark for foundation models on behavioral tasks that reveals fine-tuned behavioral models outperform general models on distributional alignment while general models lead on individual-level accuracy.
DART is a training-free router that accepts direct answers on draft agreement and allocates thinking budgets via draft entropy on disagreement, reporting accuracy gains and token reductions on math and code benchmarks across model scales.
KBSpec maintains an evolving knowledge base combining external docs and internal verifier feedback to improve LLM generation of verifiable JML specifications, achieving 10-25% higher verification pass rates.
citing papers explorer
-
StreamKL: Fast and Memory-Efficient KL Divergence for Boosting Attention Distillation
StreamKL is the first fused GPU primitive for attention KL divergence that reduces memory from O(N_Q N_K) to O(1) via an online one-pass formulation and tile-wise recomputation.
-
LoHoSearch: Benchmarking Long-Horizon Search Agents Beyond the Human Difficulty Ceiling
LoHoSearch is a new benchmark of 544 KG-constructed questions across 11 domains where the strongest search agent scores 34.74% and context strategies add at most 6.8%.
-
LITMUS: Benchmarking Behavioral Jailbreaks of LLM Agents in Real OS Environments
LITMUS is the first benchmark using semantic-physical dual verification and OS state rollback to measure behavioral jailbreaks in LLM agents, revealing that even strong models execute 40%+ of high-risk operations and exhibit execution hallucination.
-
Agent-ValueBench: A Comprehensive Benchmark for Evaluating Agent Values
Agent-ValueBench is the first dedicated benchmark for agent values, showing they diverge from LLM values, form a homogeneous 'Value Tide' across models, and bend under harnesses and skill steering.
-
ReLibra: Routing-Replay-Guided Load Balancing for MoE Training in Reinforcement Learning
ReLibra uses pre-known token-to-expert routing from RL rollouts to perform inter-batch expert reordering and intra-batch replication, delivering up to 1.6x higher throughput than Megatron-LM and 1.2x over oracle-equipped EPLB while staying within 6-10% of an ideal balanced baseline.
-
HarmfulSkillBench: How Do Harmful Skills Weaponize Your Agents?
Harmful skills in open agent ecosystems raise average harm scores from 0.27 to 0.76 across six LLMs by lowering refusal rates when tasks are presented via pre-installed skills.
-
HWE-Bench: Benchmarking LLM Agents on Real-World Hardware Bug Repair Tasks
HWE-Bench is the first repository-level benchmark for LLM agents on real hardware bug repair, where the best agent fixes 70.7% of 417 tasks but drops below 65% on complex SoC projects.
-
OccuBench: Evaluating AI Agents on Real-World Professional Tasks via Language Environment Simulation
OccuBench is a new benchmark for AI agents on real-world occupational tasks via LLM-driven simulators, showing no model dominates all industries, implicit faults are hardest, and larger models with more reasoning perform better.
-
AlpsBench: An LLM Personalization Benchmark for Real-Dialogue Memorization and Preference Alignment
AlpsBench supplies 2500 real-dialogue sequences with verified memories to benchmark LLM extraction, updating, retrieval, and utilization of personalized information.
-
MentalHospital: A Virtual Environment for Evaluating Psychiatric Clinical Encounters
EHR-derived standardized patients and dual-track evaluation reveal LLMs trail clinicians by 37.28 points on full psychiatric encounters, with mental-status assessment the main bottleneck.
-
Hallucination Self-Play: Bootstrapping Reinforced Detector via Evolved Generator
Hallucination Self-Play co-evolves a generator and detector from one base LLM via RLAIF and RLVR, lifting a 7B model to match advanced LLMs on RAGTruth faithfulness detection.
-
Lynx: Progressive Speculative Quantization for accelerating KV Transfer in Long-Context Inference
Lynx partitions KV cache bits into anchor and residual streams for progressive transfer, enabling speculative decoding on partial data followed by verification to match BF16 accuracy at 4-bit-like TTFT.
-
Can Agents Generalize to the Open World? Unveiling the Fragility of Static Training in Tool Use
Static SFT and RL training for tool-use agents leads to performance drops under open-world distributional shifts across perception, interaction, reasoning and internalization; perturbation-augmented fine-tuning is proposed as mitigation.
-
Self-GC: Self-Governing Context for Long-Horizon LLM Agents
Self-GC governs agent context as indexed objects with planner-proposed actions, achieving 84.85% no-impact on future continuations on a hard set versus 54-70% for baselines.
-
SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering
SmoothAgent introduces lookahead context engineering to eliminate transformation overhead in LLM agents, reducing TTFT by up to 11.9x through proactive KV cache preparation.
-
Moral Safety in LLMs: Exposing Performative Compliance with Puzzled Cues
Hiding explicit demographic labels raises harmful LLM decisions by +4.4 pp even when models correctly recover the identity, exposing performative rather than genuine moral safety.
-
OmniCoT: A Benchmark for Global and Multi-Step Panoramic Reasoning
OmniCoT is a new panoramic reasoning benchmark with 6.7K eval, 1K real, and 14.3K training examples plus a two-stage SFT+GRPO training method to enforce global 360-degree consistency.
-
CORTEX: High-Quality Cross-Domain Organization of Web-Scale Corpora through Ontological Corpus Graph
Cortex uses an Ontological Corpus Graph to structure web-scale corpora, creating a refined 24.14B-token corpus and a new benchmark validated on eight LLMs.
-
SpreadsheetBench 2: Evaluating Agents on End-to-End Business Spreadsheet Workflows
SpreadsheetBench 2 provides 321 expert-validated tasks from authentic business data showing frontier LLMs reach only 34.89% overall accuracy on end-to-end spreadsheet workflows.
-
Dockerless: Environment-Free Program Verifier for Coding Agents
Dockerless uses agentic repository exploration to verify patches without execution, enabling SFT and RL training of coding agents that reach 62.0/50.0/35.2% resolve rates on SWE-bench Verified/Multilingual/Pro while matching environment-based results.
-
Agentic Generation of AST Transformation Rules for Fixing Breaking Updates
BigBag generates reusable AST transformation rules via LLMs that achieve 78.6% fix rate on 157 breaking dependency updates and 33.3% cross-project transfer overall.
-
BehaviorBench: Benchmarking Foundation Models for Behavioral Science Tasks
BehaviorBench is a benchmark for foundation models on behavioral tasks that reveals fine-tuned behavioral models outperform general models on distributional alignment while general models lead on individual-level accuracy.
-
DART: Draft-Agreement Routing for Training-Free Adaptive Thinking Budgets in Hybrid Reasoning Models
DART is a training-free router that accepts direct answers on draft agreement and allocates thinking budgets via draft entropy on disagreement, reporting accuracy gains and token reductions on math and code benchmarks across model scales.
-
KBSpec: LLM-driven Formal Specification Generation with Evolving Domain Knowledge Base
KBSpec maintains an evolving knowledge base combining external docs and internal verifier feedback to improve LLM generation of verifiable JML specifications, achieving 10-25% higher verification pass rates.
-
A Systematic Evaluation of Black-Box Uncertainty Estimation Methods for Large Language Models
A unified benchmark of 24 black-box UE methods for LLMs finds no universal winner but favors methods that reason over answer candidates and hybrid combinations of signals.
-
LegalWorld: A Life-Cycle Interactive Environment for Legal Agents
LegalWorld is a life-cycle interactive environment modeling Chinese civil litigation as five causally connected stages grounded in 75,309 judgments, paired with LongJud-Bench for cross-stage agent evaluation.
-
Models Take Notes at Prefill: KV Cache Can Be Editable and Composable
KV caches function as notebooks of prefilled conclusions, enabling field-level edits that recover decisions (especially with CoT) and position-portable skill composition with near-identical outputs at O(L) cost.
-
From Shield to Target: Denial-of-Service Attacks on LLM-Based Agent Guardrails
Attackers can force LLM guardrails into extended reasoning loops via optimized payloads, causing 13-63x token amplification and up to 148x latency in agent systems.
-
Which Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMs
ModSleuth reconstructs dependency graphs from public artifacts for four LLM releases, recovering 1,060 source-verified dependencies and exposing license issues, train-evaluation coupling, and documentation gaps.
-
FORT-Searcher: Synthesizing Shortcut-Resistant Search Tasks for Training Deep Search Agents
FORT synthesizes shortcut-resistant search tasks by controlling four identified shortcut risks across entity selection, graph construction, question formulation, and refinement, producing training data that yields agents with longer search trajectories and top performance among open-source models on
-
The Arbiter Agent: Continually Monitoring Multi-Agent Conversations to Detect Emergent Misalignment
Introduces the Arbiter agent for budget-constrained real-time detection of emergent misalignment in multi-agent conversations, with evaluations showing reliable early detection aided by active inspection tools.
-
RealMath-Eval: Why SOTA Judges Struggle with Real Human Reasoning
RealMath-Eval benchmark shows LLM judges have an evaluation gap, performing worse on diverse real human math reasoning than on synthetic solutions due to greater error diversity and higher surprisal.
-
Experience Makes Skillful: Enabling Generalizable Medical Agent Reasoning via Self-Evolving Skill Memory
SkeMex distills agent trajectories into value-aware skills organized in general/task/action branches and evolves them via a closed-loop Read-Write-Assess-Govern process, outperforming prior memory agents on clinical tasks.
-
Data Agents Under Attack: Vulnerabilities in LLM-Driven Analytical Systems
The paper introduces a layered vulnerability framework and attack taxonomy for LLM-driven data agents and demonstrates attacks on four open-source and two production systems.
-
AudioProcessBench: Benchmark for Identifying Process Errors in Audio-Grounded Reasoning
AudioProcessBench is a new benchmark with segmented and annotated reasoning traces from six audio and omni-language models for step correctness identification and error-type detection in audio-grounded reasoning.
-
WhiFlash: Accelerating Speculative Decoding with Token-Level Cross-Paradigm Routing
WhiFlash introduces token-level cross-paradigm routing between autoregressive and diffusion drafting models, with cache optimizations, to raise acceptance lengths and deliver up to 69.6% throughput gains over EAGLE-3.
-
Knowledge Index of Noah's Ark
Introduces KINA benchmark with 899 items over 261 disciplines, formal (1-1/e) coverage guarantee and bonus-on-bar tournament theorem, plus evaluations of 42 models with top score 53.17%.
-
ClinicalMC: A Benchmark for Multi-Course Clinical Decision-Making with Large Language Models
ClinicalMC is a benchmark of 1,275 Chinese and 5,804 English multi-course clinical samples across four stages, evaluated via a multi-agent framework on closed-source, open-source, and medical LLMs in static and dynamic settings.
-
CultureForest: Understanding and Evaluating Cultural Norm Grounded Reasoning in LLMs
CultureForest benchmark shows top LLMs degrade sharply on open-ended cultural reasoning tasks, exhibit regional disparities, and are limited more by effective use of knowledge than by lack of knowledge itself.
-
WorldCoder-Bench: Benchmarking Physically Grounded 3D World Synthesis
Introduces WorldCoder-Bench and StateProbe for evaluating LLM-generated physically grounded 3D browser worlds, with frontier models reaching at most 27.8% verification coverage.
-
Move the Query, Not the Cache: Characterizing Cross-Instance Latent Attention Redistribution Across GPU Fabrics
On a real multi-node H100 cluster the authors show that for MLA, routing the ~1 KB compressed query row is cheaper than moving cache chunks and supply a topology-aware cost model accurate to ~7% on IBGDA fabrics.
-
The Invitation Trap: Proactive Availability Backdoor in LLMs via Conversational Induction
The paper presents Proactive Availability Backdoor (PAB) attacks on LLMs that achieve 73.1% effective success rate by proactively inducing users via suggestions in a Five-Factor Model simulation.
-
Every Act Has Its Price: Compressed Moral Composition in Frontier LLMs
Moral Trolley Arena shows frontier LLMs produce composite moral preferences that are compressed rather than additive functions of calibrated component act strengths across Moral Foundations Theory.
-
LiveBrowseComp: Are Search Agents Searching, or Just Verifying What They Already Know?
LiveBrowseComp shows search agents rely on intrinsic knowledge on standard benchmarks, with scores dropping 25-40 points and closed-book accuracy below 2% on questions about facts from the prior 90 days.
-
VitaBench 2.0: Evaluating Personalized and Proactive Agents in Long-Term User Interactions
VitaBench 2.0 introduces a benchmark for long-term personalized and proactive agent behavior, with results indicating substantial gaps in current frontier LLMs.
-
AssertLLM2: A Comprehensive LLM Benchmark for Assertion Generation from Design Specifications
AssertLLM2 introduces a benchmark of 83 designs supporting bug-prevention and bug-hunting assertion generation tasks with evaluation across syntactic, formal, coverage, and mutation-based metrics.
-
FTibSuite: A Comprehensive Resource Suite for Tibetan Vision-Language Modeling
FTibSuite provides human-verified multimodal corpora, Tibetan-adapted benchmarks with quality controls, and a baseline VLM showing gains on tasks like MMBench while preserving Chinese capabilities.
-
RepoMirage: Probing Repository Context Reasoning in Code Agents with Perturbations
RepoMirage uses semantics-preserving perturbations on SWE-Bench to show code agents lack repository context reasoning, with performance falling sharply on extended structure tasks, and introduces RepoAnchor as a structure-first fix.
-
Not only where, But when: Temporal Scheduling for RLVR
Temporal scheduling of credit allocation criteria over RLVR training, using trajectory percentiles to target heterogeneous behaviors, yields more stable policy entropy and better reasoning benchmark results than static allocation.
-
Towards Reliable Fetal Ultrasound Interpretation with Multi-Agent Collaboration
FetUSAgents uses tool-augmented multi-agent collaboration and Dual-Path Evidence Arbitration to exceed prior MLLMs by over 25% on a new fetal ultrasound VQA benchmark.